I'm comparing cpu usage and latency between the whalin (2.0.1) and sallings
(2.3.1) java clients.
(I'm only considering for synchronous operations).
I'm interested in the relative performance between the clients, not the absolute
numbers.
My benchmark is 1M get operations for a single key spread across 5 threads with
6 memcached servers.
The clients are running on a 4-core box. The memecached processes are running
on a separate 4-core box.
For cpu consumption, the sallings client consumed ~25% less user cpu and ~60%
less system cpu.
For latency, the sallings client had ~10% greater average latency.
The cpu results are at least directionally consistent with another comparison
here,
http://blogs.sun.com/elambert/entry/memcached_java_client_performance_on
The latency results are not ... any guidance on what might have caused the
sallings client
to have greater latency?
thanks,
Miguel
